SHARING VEHICLES & SHARING RIDES IN REAL TIME: OPPORTUNITIES FOR

SELF-DRIVING FLEETS

Krishna Murthy Gurumurthy

Department of Civil, Architectural and Environmental Engineering

The University of Texas at Austin

gkmurthy10@utexas.edu

Kara M. Kockelman, Ph.D., P.E.

(Corresponding Author)

Professor and DeWitt Greer Centennial Professor of Transportation Engineering

Department of Civil, Architectural and Environmental Engineering

The University of Texas at Austin – 6.9 E. Cockrell Jr. Hall

Austin, TX 78712-1076

kkockelm@mail.utexas.edu

Benjamin J. Loeb

Delaware Valley Regional Planning Commission

Philadelphia, PA 19106

bloeb@dvrpc.org

Forthcoming chapter in Advances in Transport Policy and Planning: The Sharing Economy and

the Relevance for Transport

Editor: Elliot Fishman

ABSTRACT

Access to shared and fully-automated or “autonomous” vehicles (SAVs) is coming, and expected

to be popular and cost-effective, especially for city dwellers. This chapter synthesizes and

summarizes research on SAVs, including dynamic ride-sharing (en route), range-constrained

electric SAV (SAEV) operations, SAV fleet costs, and variable road pricing in a world of AVs,

where vehicle-miles traveled (VMT) rise and congestion worsen. Researchers consistently find

that a single SAV with long range and fast refueling can replace 6 or more household vehicles in

countries with high vehicle ownership, even when serving long-distance trips. That number falls a

bit when SAVs are range constrained and/or have long recharging times. Zero-occupancy VMT,

called empty VMT, will be a problem for urban-network congestion levels if travelers do not share

rides with strangers (increasing average vehicle occupancy) and road tolls are not included.

Expected costs are consistently under USD $0.75 per revenue-mile, assuming the self-driving

technology add USD $25,000 or less to conventional vehicles. SAEVs will bring some emission

savings when vehicle-to-grid consumption is managed well. Fleet sizing, fare economics and

regional policies will be key to managing this SAV future.

Keywords: Shared mobility, autonomous, shared autonomous vehicles, dynamic ride-sharing,

electric vehicles, future transport, pricing

BACKGROUND

Fully-automated or autonomous vehicles (AVs), referred to as Level 4 AVs according to SAE

International1, are in an advanced testing phase in the U.S. states of Arizona, California,

Pennsylvania, and Texas, among others. It will be no surprise if AVs are widely available for

public use in the next five years. Technology companies, like Waymo (AV subsidiary of Alphabet

Inc.) and Uber (Transportation Network Company, or TNC), are spearheading innovation in AV

technology alongside original equipment manufacturers (OEMs), and significant progress has

been made in as little as the last decade. Boasting benefits in several realms, such as travel time,

energy and safety, AVs are geared to disrupt the market in the next decade (Kockelman et al., 2016

& 2018). The benefits they have to offer, however, do come with a high price-tag. As in any

industry, the cost of innovation is eventually burdened on the consumer, and, consequently, AVs

are expected to be expensive to own, at least in the early stages of deployment. This chapter focuses

on a particular near-term, and potential long-term, alternative to AV ownership – shared AVs

(SAVs).

A fleet of SAVs operated by one entity and shared by a community can be used to make trips and

operate with minimal human intervention. One central control-unit will likely be sufficient to

perform all operations for the entire fleet, thereby, significantly cutting down the required human

capital. SAVs are also expected to be more profitable than current forms of car sharing and ride-

hailing services, largely owing to the absence of driver-related costs and human error (Fagnant and

Kockelman, 2015). The improved safety of AVs will also mean that operators may pay lower

insurance premiums, and this will allow them the flexibility to charge lower fares whilst having a

similar or better profit margin as compared to today’s ride-hailing services (Clements and

Kockelman, 2017). Aside from the profitability of the fleet from an operator’s perspective, they

will also be a convenient yet affordable, door-to-door service for everyone. Nowadays, people tend

to use their personal vehicles inefficiently, i.e., with several empty seats and for only about one

hour a day, on average (Fagnant and Kockelman, 2018). Personal vehicles may be the most

convenient alternative now, but they may start to seem expensive in contrast to SAVs. When they

begin to make better use of these unused seats via dynamic ride-sharing (or DRS, where strangers

share rides for a part of their trip), SAVs can keep fares to a minimum. Shared usage, especially

in conjunction with cleaner technology like electric propulsion, can produce impressive

environmental benefits (Farhan and Chen, 2018).

Shared mobility is an obvious next step to utilizing AVs efficiently and research on this topic is in

its infancy. This chapter aims to review the different dimensions of studies previously completed

on this topic and provides an anticipated trajectory for future research. Some of the key elements

covered here are the operational aspects of SAV use, travel demand for shared mobility, costs of

these services, impact of electrification, and policies to better accommodate this evolving option.

Research on SAVs has been a hot topic for less than a decade, but numerous researchers from

around the globe have generated many valuable articles on different facets of SAV use. In terms

of nomenclature used in this chapter most researchers use the term ‘SAV’ for 4-seater vehicles,

1 SAE demarcates 5 levels of autonomy, with level 4 and 5 being fully-autonomous and the other lower levels

needing human input. Level 5 differs from 4 in the AVs ability to drive off-road and without connectivity or

feedback from the infrastructure (https://saemobilus.sae.org/content/j3016_201609).

and aBuses for with larger-capacity shared autonomous vehicles. Terms like ‘Autonomous

Mobility-on-Demand’ (AMoD), ‘Shared use Automated Mobility Services’ (SAMS), and

‘Autonomous Taxis’ (ATs or aTaxis) are also used. The following sections detail different

dimensions of SAV and aBus research, ending with descriptions of literature gaps worthy of

further exploration.

FROM SHARING VEHICLES TO SHARING RIDES

ZipCar began its operation in early 2000 and was key in popularizing carsharing. The growing

popularity of AVs due to the DARPA challenge (Buehler et al., 2009) encouraged several

researchers to delve into a mixture of AVs and carsharing. Burns et al. (2013) used aggregated trip

data for three U.S. cities (Ann Arbor, Michigan; Babcock Ranch, Florida; Manhattan, New York)

to solve, both analytically (using pre-defined relationship to system metrics) and using simulation,

for an approximate fleet requirement. Their results suggested that a fleet range of 2,000 to 18,000

SAVs were required depending on the region’s trip density and total number of trips in peak period.

Similarly, fleet utilization varied between 70-90% in peak periods and had reasonable response

times. Although their research made significant assumptions (like uniform origin-destination [O-

D] distribution) to simplify the process of estimation, it was one of the first studies on SAVs,

quantifying metrics like empty driving and customer wait times. Some of these shortcomings were

addressed by Spieser et al. (2014) in their study for Singapore using a more robust methodology.

Using real trip data from a travel survey fused with network statistics from a taxi dataset, they

showed that Singapore’s travel demand could be satisfied with just one-third of the then existing

vehicle fleet. Both these studies strongly recommended the transition to SAVs, but also advised

caution on the increase in vehicle-miles traveled (VMT) that can result from unoccupied travel.

Fagnant and Kockelman (2014) were able to quantify this rise in VMT at 11% from studying travel

patterns on a grid representing downtown Austin, Texas. A higher replacement rate of 11 to 1 (i.e.,

the number of conventional vehicles that can be replaced by 1 SAV) was observed in this study,

which is nearly four times as much as Spieser et al.

Brownell and Kornhauser (2014) explored the used of SAVs at a much larger level. Travel

behavior for the entire state of New Jersey was aggregated in grids, arising from satellite feed

pixels, due to the lack of a detailed road network. Their analysis tested SAVs as a paratransit

service, and primarily showed that allowing O-D aggregation for pickup and dropoff (rather than

fixed stops) increased average vehicle occupancy (AVO), i.e., number of travelers accommodated

in one vehicle, due to increased convenience and comfort from closer pickup and dropoff points.

Similar to Spieser et al. (2014), the entire state’s travel demand could be served by one-third of

the personally-owned fleet, but it is important to note that Brownell and Kornhauser (2014)

simulated a preliminary version of DRS. Burghout et al. (2015) was able to use a detailed network

for Stockholm, Sweden but with a limitation of only zone level O-D available for personal-vehicle

trips. They were able to test three subsets of DRS matched at the zone level, i.e., trips with common

O-D pairs, trips with common O’s but different D’s, and trips with different O’s but common D’s.

Owing to zone level information, intra-zonal travel time was assumed, and travel times were fixed

at a fraction of the free-flow speed. An overall reduction in VMT was observed, thanks to DRS,

with a fleet just 5% the size of the private fleet serving all trips. Although a version of DRS was

tested by Brownell and Kornhauser (2014) and Burghout et al. (2015), severe aggregation and use

of static travel times may have led to optimistic results of AVO and replacement rate (in hindsight)

for the scenarios tested. It was common consensus by now that realistic behavior using dynamic

traffic assignment (DTA), i.e., a simulation-based analysis, was required to understand the impact

of an SAV fleet, and especially that of DRS.

Large-Scale Agent-Based Simulations

Simulation-based studies tried to incorporate realistic behavior but realism was added step by step

over time, beginning with understanding an SAV fleet’s operation without DRS. The increased

need for realism in travel demand modeling, in terms of both spatial and temporal scales, resulted

in the development of an agent-based model called the multi-agent transport simulation, or

MATSim (Horni et al., 2016), and it quickly became popular for its ability to run DTA on a large

scale. Fagnant et al. (2015) simulated SAVs on an idealized grid-network, of longitudinal and

lateral links of equal length, representing downtown Austin, Texas, and using hour-by-hour travel

times from MATSim. A sample of trips from the region’s travel demand model was simulated to

use SAVs to account for uncertainty of the initial market toward AVs, while also spatially

constraining the SAV operation to the downtown area. Their 24-hr weekday simulation found that

each SAV could make all the trips originally made by nine conventional vehicles, and this was

much higher than the replacement rate of 3 to 1 found in Spieser et al. but much lower than the 20

to 1 estimated by Burghout et al. This was extraordinary considering that Austin’s population was

only one-fifth that of Singapore in 2015, i.e., Austin likely had a much lower trip density, but it

may also have stemmed from their downtown comparison, as opposed to Spieser et al.’s study of

the entire region. However, Fagnant et al. also found an estimated 8% increase in VMT from SAVs

operating unoccupied. On the bright side, their results did show better utilization of the fleet, and,

therefore, possibility of reduced emissions from fewer cold starts and faster turnover of the fleet.

Bösch et al. (2016) also conducted a similar analysis, using demand generated for Zurich,

Switzerland in MATSim, running SAV simulations outside the MATSim environment. Their

results of a 10 to 1 vehicle replacement ratio validated Fagnant et al.’s high rate, even when as

little as 10% of the demand was served from an assumed low market penetration. From their 1000

scenarios simulated, they were able to show that fleet size was a function of coverage area rather

than the demand served. In both these MATSim-related simulations, the main drawback was that

SAVs were not simulated internally with congestion feedback.

Taking the next step in SAV simulations, Bischoff and Maciejewski (2016) integrated a

dynamically-responsive SAV fleet within MATSim to observe congestion effects. Their work

simulated 100,000 SAVs for Berlin, serving 2.5 million trips, and observed a 10 to 1 replacement

rate for conventional vehicles, thus supporting the simplifications made by Fagnant et al. and

Bösch et al. (2016). Results from their simulation showed increase in drive time (up to 17%) as

compared to before the SAV service, which can be a proxy for VMT increase as observed by

Fagnant et al., but no significant delays were seen. Although absence of significant congestion was

attributed to better flow by SAVs potentially coordinating on the roadway, this is unlikely in the

near-term because of safety concerns and human drivers fearing tight headways to SAVs, at least

initially.

Simulating Dynamic Ride-Sharing

An increase in VMT was a common conclusion in the literature at this point which would

eventually lead to increased congestion, despite several personally-owned vehicles being replaced.

With prior research supporting the benefits of DRS, albeit with limitations, a natural next step was

to observe network-wide impacts when an SAV service is operated with DRS. Fagnant et al.'s

(2015) model on the gridded-network for Austin, Texas was extended by Fagnant and Kockelman

(2018) to include DRS capabilities. Several conditions for added travel time to riders were included

to filter away unrealistic ride-matching (with travelers experiencing high wait times or added

delays) and an overall increase in the SAV fleet’s serviceability was observed. This study, which

was published 2016, showed that the 8% increase in VMT that was previously observed, was at

least partially mitigated, even with only a sample of the population using DRS. Adding more

flexibility to the ride-matching procedure (like allowing greater than a 40% increase in travel time

from DRS as opposed to direct O-D travel time, but with wait time constrained to approximately

10 min) capped rising VMT at 1%. When more travelers (11% of the region’s demand) where

simulated to use the service, the overall VMT decreased. DRS was, therefore, seen as a potential

solution to the rising congestion.

Maciejewski et al. (2017) documented ride-matching and traffic flow related algorithms used by

Bischoff and Maciejewski, and introduced modifications to MATSim (specifically, by

programmatically solving the dynamic vehicle routing problem) that allowed them to simulate

DRS. This was further enhanced by Hörl (2017) further, where hetested how an SAV fleet behaved

when rides were shared versus solo travel. By simulating a fleet where half the vehicles explicitly

offered DRS while the other half offered a personal ride, Hörl was able to show that DRS may be

preferred at off-peaks due to the low cost of a shared ride, together with DRS having low demand

that reduced chances of matching, and, consequently, added delay. Personal rides were preferred

during the peak to benefit from low travel times. Using the popular Sioux Falls test network, Hörl

also simulated the fleet twice, once offering DRS and once offering only personal rides, to discuss

efficient seat usage and attained a 1.64 maximum AVO with DRS, but was unable to comment on

VMT because of the artificial network used. Better utilization of empty seats may be seen as a

proxy for curbing rising VMT, but route deviations arising from DRS may have outweighed the

saved VMT.

Around the same time, Levin et al. (2017) developed a similar DRS application in a simulator that

incorporated more realistic traffic flow for their case study of Austin, Texas. Their study provided

a more measured assessment of AVs benefits, as opposed to more optimistic results from past

studies that did not take into account congestion effects using realistic flow models. Shorter travel

times and reduction in VMT was observed only when a small fleet of around 2,000 SAVs served

nearly 63,000 trips in Austin’s CBD with DRS for 2 hours in the AM peak. Increasing fleet size

decreased the necessity to share rides and decreased fleet utilization (i.e., many SAVs were idle).

Each SAV was serving, on average, 31 trips in one day (each trip averaging 2.3 mi), which is

comparable to that observed by Fagnant and Kockelman (2018) in their simulation with 11% of

the region’s demand.

Up until now, studies focused primarily on SAV use, be it with or without DRS. Martinez and

Viegas (2017) studied a fleet of SAVs operating in Lisbon, Portugal, and included mode choice in

their simulation. Two types of shared vehicles were available to choose from: 4-seater AVs and

either 8- or 16-seater aBus, in addition to walk, or subway as available modes with the choice

made based on an estimated model (substituting taxi and buses for their AV alternative). As a first

in the literature, overall VMT was reduced by an astonishing 30% in their study when SAVs and

aBuses were available, with results indicating significant emission benefits as well. A large AVO

of 2.0 in SAVs and 4.2 in aBuses is likely the most significant contributor to this large decrease in

VMT. This combined fleet of SAVs and aBuses could replace nearly 10 conventional vehicles.

Another study to include mode choice was Liu et al. (2017) who also studied SAV fare impacts,

but without DRS. They modeled choice between conventional vehicles, transit and SAVs, and

served requests using an SAV, with a wait time threshold of 10 minutes. Although they observed

significant energy and emission savings by considering better drive cycles for AVs (Kockelman

and Boyles, 2018), the increase in VMT from the absence of DRS, and resulting congestion, was

found at all fare levels. However, this will depend on how AVs are priced in the future. Further,

the authors highlighted the importance of modeling destination choices for future studies to capture

changes in trip distribution whilst using low-cost SAVs. Heilig et al. (2017) solved this problem

with a case study for Stuttgart, Germany. Their model allowed choices for walking, biking, transit

and sharing rides in SAVs, along with destination choice. They found overall VMT in Stuttgart

reduced by about 20% using a relatively small fleet, just 15% the size of the existing private vehicle

fleet, but similar to Bösch et al.'s (2016) findings who did not simulate DRS. Interestingly, the

walk, bike and transit modal shares in Liu et al.’s future scenario increased, similar to Heilig et al.

(2017) , although one might assume that low-cost alternatives of 50¢/mi for SAVs would prevent

this. Zhao and Kockelman (2018), used a more familiar approach of a four-step demand model,

and tested fare sensitivities when destination choice was modeled. Spanning 9 scenarios, they

showed that VMT increased under different operating costs, parking costs, tolls.

Operational Nuances, Parking and Solving the First-Mile-Last-Mile Problem

The SAV literature’s focus eventually shifted from direct applications of SAV and DRS operations

to understanding more nuanced information relating to concepts like proactive vehicle relocations

and ride-matching processes, as noted next, as well as parking demand changes and first-mile-last-

mile (FMLM) benefits.

An SAV fleet’s operational efficiency and ride-matching rate were tested by Hörl et al. (2018) and

Ruch et al. (2018) using a common framework. The studies focused on testing two ride-matching

algorithms (a simple heuristics and a bipartite optimization routine) and two rebalancing

techniques for case studies of Zurich and San Francisco using MATSim’s traffic assignment

module. Their results suggested that smart matching and rebalancing algorithms are able to quickly

match travelers to their SAVs, and can give a competitive advantage in the market when several

companies are competing for demand. However, from a single operator perspective, reduction in

wait times proved to be beneficial. Hyland and Mahmassani (2018) tested six distinct dispatch

strategies for an SAV fleet without DRS simulated using the Manhattan, New York grid. Their

study revealed that any future SAV fleet is better off with an optimization framework in order to

either reduce fleet size or empty VMT, and that this also depends on the spatial structure (e.g.,

dense or sprawling) of the region. Their research helps inform initial SAV operators dispatch

strategies based on the region and their objective.

Large availability of travel data from cellphone records, prompted Gurumurthy and Kockelman

(2018) to conduct a benchmark test for DRS. Their study, based on travel in Orlando, Florida,

generated the region’s demand and simulated them as served by an SAV fleet with DRS. The

initial aggregate dataset was discretized temporally and spatially and DRS-based data analytics

showed that around 60% of single-occupant trips in the region could be served by 60,000 SAVs,

or 25 travelers per vehicle. This is lower than the 31 person-trips served per SAV in Fagnant and

Kockelman (2018) but the discrepancy arose mainly from the sprawling region and special nature

of recreational trips in Orlando. This study validated Hyland and Mahmassani's (2018) hypothesis

on spatial structure with around 3% added VMT when ride-matches were maximized. However,

it is important to note that even under sprawling conditions, DRS was found to be viable.

SAVs are expected to lower parking demands, at least in downtown areas and other big trip

attractors (like airports and hospitals). However, very few studies have analyzed parking demand

changes. Using a simple simulation model, Zhang et al. (2015) predicted an optimistic 90%

decrease in parking demand on a 10-mile hypothetical grid with only 2% of the population using

SAVs, and even greater parking-demand reductions based on the willingness to share rides. Two

years later, using a model that allowed discrete-time-based simulations, Zhang and Guhathakurta

(2017) estimate a 5% less parking land required across Atlanta, thanks to just a 5% of vehicle

owners shifting to SAVs. Moreover, parking demand was estimated to shift away from downtown

spaces to low-income areas, if parking charges are assessed and privately owned AVs are allowed

to drive empty in search of lower-cost parking. More recently, Millard-Ball (2019) focused on

parking demand using travel data from the San Francisco demand models and tested how parking

behavior affected the system, by comparing parking as before, free-for-all parking, SAVs returning

home, and SAVs cruising around. He concludes that a shift to SAVs would not be beneficial

overall, due to more driving and added congestion, and recommends using a time-based charge for

SAV usage, irrespective of whether it is parked or moving. In a region’s CBD, this may be useful

to apply across the board.

Shared AVs with DRS, self-driving minibuses, and other versions of aBuses can also serve in lieu

of existing buses, in addition to delivering travelers to and from train stations, much like present-

day TNCs offering a lower cost ride-sharing alternative. Liang et al. (2016) considered SAV

service for a train station in Delft, Netherlands. FMLM service was designed using profit-

maximization strategies that served all requests. This zone level study concluded that the FMLM

service can be profitable, though this study used fixed demand. Average VMT increased per

vehicle with small fleet sizes, but no comment was made on network impacts. Scheltes and de

Almeida Correia (2017) extended that work, to analyze network metrics via simulations that reflect

additional SAV nuances (like dispatch and relocation) and test the impact of an SAV’s travel

speed. Relocation strategies increased response times, but even with FMLM, increased overall

VMT and added congestion. Shen et al. (2018) included DRS in their FMLM study and focused

on when and where SAVs can best complement existing transit lines. They conclude that replacing

underutilized bus lines or other forms of transit with SAVs will be beneficial. Farhan et al. (2018)

sought to optimize SAV operations via DRS and were able to achieve a VMT reduction of nearly

37%, versus without using DRS in Seattle. Pinto et al. (2018) used a dynamic assignment model

to incorporate many details of transit use, such as boarding and alighting times and queue

formation based on available space within transit vehicles. They analyzed how FMLM-serving

SAVs can benefit such systems, and found that SAV vs. transit fares impacted mode choice

significantly. Their bi-level solution framework showed that, overall, SAVs are expected to be a

viable option for FMLM use (especially with DRS), in terms of network benefits.

ENERGY AND EMISSIONS

Utilizing the SAV fleet better than existing personal vehicle fleet will lead to some emission

improvements. Fagnant and Kockelman (2014) showed that reduction in cold and warm starts (to

0.05 and 0.7, respectively, per person trip) from shifting to an SAV fleet can reduce CO2 emissions

and particulate matter in the air. These results stayed true even with a better simulation framework

by Fagnant et al. (2015) showing an 85% reduction in cold starts. SAVs and aBuses offering DRS

as in Martinez and Viegas (2017) were able to reduce CO2 emissions by almost 40% across all

road modes.

Electrification can mitigate some of the issues caused by SAVs with internal combustion engines.

These include eliminating engine idling, lower emissions to counteract added VMT and a quieter

experience for the customer. EVs are also well suited for the ultra-rigorous use of an SAV fleet

with likely less maintenance needs without the complexity of an ICE powertrain. The idea of

electrified SAVs (or SAEVs) has gained the interest of researchers lately with many conducting

agent-based models to understand how these systems operate. Researchers’ goal is generally to

determine the feasibility of electrifying an SAV fleet, so the first step is looking at system

performance. Gasoline vehicles’ nearly immediate refuel times and ubiquitous refueling stations

make range a non-issue, but for EVs, recouping range can be a concern. In the near future, cities

may have the power grid and charging infrastructure needed to provide inconsequential charge

times, but in the meantime it is not obvious if SAEVs can deliver timely, on-demand transportation

service with today’s technology. Chen et al. (2016), Kang et al. (2016), Yang et al. (2017) and

Loeb et al. (2018), all simulate SAEV fleets, focusing on response times and necessary fleet sizes.

These papers published between 2016 and 2018 have modeled SAEV fleets with increasing

complexity.

Chen et al. (2016) modeled SAEVs in a generic grid imitating a large, metropolitan area with a

dense, urban core. Trips were generated based on their proximity to the center of the area. The

fleet and charging stations were generated based on demand through a “warm start” period to

encourage reasonable vehicle start locations and charging station placement. Their goals were to

understand how charge time and vehicle range affected the fleet size and the number of charging

stations needed to serve demand quickly. They found a fleet of 200-mile range vehicles could be

20% smaller than one composed of vehicles with 80-mile range and 30-minute fast-charging

reduced fleet size by 30% compared to 4-hour charge times. Combining fast-charging and long

range gave a 44% decrease in fleet compared to lower ranges and slower charging. The number of

charging stations needed to meet demand did not vary much based on these modeling settings, but

the number of chargers at each station could be cut by 45% when upgrading from slow charging

to fast charging for the short range case and cut by 86% for the long-range case. Vehicle response

times for trip requests was 7 - 10 min on average, unoccupied travel accounted for 10 - 14% of

SAEV VMT (a very small portion of which was for charging) and each vehicle could replace up

to 6.8 privately owned vehicles based on daily trip-making rates.

Yang et al. (2017) used a more specific and detailed simulation environment modeled after

Shanghai China. They collected a month’s worth of taxi travel behavior data from 13,761 vehicles,

over 23% of Shanghai’s taxi fleet. This was used to map Shanghai into a grid, and trips were

generated in each tile according to collected data (similar to methods used by Brownell and

Kornhauser, 2014). Vehicles in the simulation environment were dispatched in an unusual first-

come, last-served system, where vehicles that ended a trip most recently were dispatched first.

This was in order to reduce vehicle idle time. Several vehicle ranges were tested, varying from 93

to 217 miles. 73% of vehicles were able to conduct a full day’s taxi operations without charging

when equipped with 217-mile range. Full-load ratio (the ratio of occupied VMT to total VMT)

was 89% in the simulation compared to 67% in (non-autonomous) observed data. They calculated

that the SAEVs could reduce the total taxi fleet size by 41%, but each SAEV idled for an average

of 15 hours a day, which means the replacement rate could likely be higher.

Kang et al. (2016) took a slightly different approach by modeling an autonomous car-sharing

service, where vehicles perform charging trips and meet passengers autonomously, but are human-

driven when occupied by a user so as to more seamlessly emulate the experience of a traditional

car-sharing service. Rather than simply borrow trip data from literature, Kang et al. estimate

demand through an iterative process with a profit-maximizing objective. Their model is an in an

11 × 11 mile square simulating Ann Arbor, Michigan. Similar to Chen et al.’s findings, they

determined that metrics like wait time are not effected by vehicle range and charge times so long

as the vehicle fleet is sufficiently large. Charging stations were placed throughout the region at

locations which minimize the average distance needed to travel to them. They assumed 30 minutes

were needed to charge a 24 kWh (80-mile) battery to 80%. They found reasonable wait times of

11.9 min, a bit slower than 9.7 minutes for a non-electrified fleet.

Farhan and Chen (2018) incorporated DRS into Chen et al.'s (2016) model with an optimization

framework to obtain best matches. Their results showed that fleet sizes could be further decreased

with DRS, and DRS was effective even when vehicles fit only two travelers. The number of

charging stations required also decreased with a net increase in benefits, but a drop in VMT was

not confirmed.

Loeb et al. (2018), inspired by Chen et al. (2016), used a detailed SAEV simulation environment,

modeling the Austin, Texas region using a true network and zonal data as supplied by the local

metropolitan planning organization. Trips on the network were modeled with state-of-the-art

activity based modeling and dynamic traffic assignment in MATSim. Many concepts were

borrowed from Chen et al., such as generating charging stations based on trip demand and testing

a series of charging speeds and vehicle ranges. Unlike Chen et al. (2016), however, fleet sizes were

fixed for each scenario and many more scenarios were tested. They found that response time was

heavily dependent on charging speeds, going from under 4 min at 30-min charge times to over 9

minutes with 4-hour charge. Vehicle range did not have nearly the same impact on response times,

except with 4-hour charging time which forced response time from 9 minutes at 93-mile range to

35 minutes at 62-mile range (electric range that was fairly typical at the time but would not be

acceptable in today’s new EV market). Fleet size was by far the most impactful for response time.

A fleet size equivalent to 9 travelers per vehicle gave response times of nearly 25 minutes with 30-

min charge times and 62-mile range. Increasing the fleet size to just 7 travelers per vehicle dropped

response times to under 8 minutes for the same range and charge time. For all scenarios,

unoccupied VMT hovered around 20% of all VMT.

Lu Miaojia et al. (2018) used Ann Arbor’s travel demand model to quantify the energy use with

SAVs and SAEVs, showing that both these fleets did not significantly help with curbing energy

use or emissions compared to present-day vehicle fleets. There was an overall increase in energy

use from empty VMT, and switching to electric propulsion increased SO2 emissions (a toxic

pollutant) due to the emission intensity observed in Michigan’s electric-grid. However, the overall

energy use decreased for SAEVs as compared to gasoline-powered SAVs. Another study by Lee

and Kockelman (2019) used results from some of prior simulation studies mentioned here (Loeb

et al., 2018; Loeb and Kockelman, 2019) and supported the positive effect that SAEVs will have

in terms of energy and emissions reduction. Even though these vehicles will need to add charging

trips to their itinerary, their research show that, overall, even with pessimistic estimates, the energy

use was lower and emission reduction was higher when a fleet of SAEVs are used compared to

non-electrified vehicles.

ENABLING REAL-TIME ROAD PRICING

Futuristic AV technology enables much more than a profitable operation of a fleet of SAVs. State

of the art automation along with high-speed connectivity offered by 5G technology that is expected

to soon be available will enable better targeting of our current day road pricing techniques. As one

may anticipate, this will be especially beneficial in the near future in aiding several local and

federal authorities who are trying to mitigate rising congestion. Advanced real-time road pricing

applied along with SAVs can hold many benefits.

Sharon et al. (2017) first studied one such advanced policy which they termed delta-tolling. The

idea behind the technique is to use measures such as current travel time and free flow travel time,

that will be easily obtained using advanced 5G technology, and calculate appropriate tolls such

that the overall social welfare (arising from better mobility) of the system increases and congestion

reduces. The authors did not particularly test a system of SAVs, but this study forms a backbone

for futuristic pricing policies that are tested in other studies noted below.

The first study to jointly analyze SAVs and advanced pricing policies was Simoni et al. (2019).

The authors compared several pricing policies in two futuristic settings: one where privately-

owned AVs dominate, and the other where SAVs dominate. Their results set the precedent to

overcome increasing VMT from single-occupant SAVs, in a world where travelers are rather

unwilling to share in the near future. Two congestion pricing schemes were tested: marginal cost

pricing and a travel-time based congestion pricing. Marginal cost pricing assesses the optimal toll

depending on the delay that one additional vehicle on the link imposes on others. Travel-time based

congestion pricing set the toll depending on the time it takes to traverse a link. Both schemes help

the network get closer to system optimum. These schemes applied to a future where mode choices

between personal vehicles (both conventional and AVs), transit, walking/biking, and SAVs

existed, resulted in a 3-4% increase in overall social welfare as opposed to scenarios without

pricing. Overall congestion dropped across both the schemes and the revenue earned this way

could also be reinvested in the community. Additionally, in countries like the United States, gas

tax implemented decades ago is unable to generate enough revenue to keep up with rising

infrastructure needs. Eventually, with increase in EV usage, the amount of gas used, and,

consequently, revenue generated from the gas tax is bound to diminish. Road pricing can play an

important role as an alternate source of revenue.

With all the buzz around DRS, naturally, the next step forward from this study was to compare

how a fleet of SAVs with DRS would fare in conjunction with pricing schemes. Gurumurthy et al.

(2019) simulated competing modes in a future of SAVs where all travelers were considered

potential customers for DRS. DRS, as expected and validated by many studies described earlier,

lowered congestion compared to a fleet without DRS thanks to empty seats being put to use.

However, when Simoni et al.'s (2019) travel-time based pricing scheme was applied, more benefits

were observed. The fleet of SAVs performed better in terms of AVO (average vehicle occupancy),

revenue and overall VMT compared to scenarios without pricing policies. The study did not focus

directly into the specifics of the pricing scheme analyzed, but showed the possibility of SAVs

operating alongside such pricing policies to improve the system, even when DRS was allowed.

COSTS OF SAV SERVICE

Operational viability of a fleet of SAVs or SAEVs forms an important point of discussion in this

field of research. However, it is equally important to understand the financial aspect of these

services. Ride-hailing services by TNCs these days are not making a profit. A large percentage of

the fare revenue goes to the driver. Those that cannot turn a profit will have no market to operate,

especially without federal funding like in the case of public transit or . Burns et al. (2013) included

a cost analysis of the simplistic SAV service that they simulated. Low response times, in the order

of seconds, combined with high conventional-vehicle replacement ratios in their study showed that

SAVs were feasible at as little as $0.30 – $0.40 per mile. This is nearly one-tenth the fares that

taxis charge (about $3/mi). These results offer for more robust calculations. Spieser et al. (2014)

compared SAVs to personal vehicles (both conventional and autonomous), taking into account the

explicit yearly costs (operation, retrofitting, and maintenance included) and value of travel time.

SAVs were viable at $0.45 per mile, which was lower than both personal AVs and conventional

vehicles. Fagnant and Kockelman (2018) showed that an assumed operating cost of $0.50 per mile

would require that the fleet’s fare be assessed at about $1.00 per mile for a profitable operation

when capital costs and traveler waiting costs were included. Bösch et al. (2018a) developed a

detailed cost calculator to piece together the different contributions to the cost of an SAV service.

Their study concluded an operational cost range of $0.50 per mile, for SAVs with DRS, to $0.70

per mile for private ride in SAVs, which means that a profitable fare may be close to $1.00 per

mile that Fagnant and Kockelman (2018) predicted. Bösch et al. (2018b) further used MATSim to

run costs assessed in Bösch et al. (2018a) but across different modes, such as personal rides, shared

rides, and transit. They established the tradeoff between VMT and total delay arising from these

services, and suggested that the high costs associated with increased VMT can still be offset by

gains from reduced travel times observed with SAV services. This is assuming that AV technology

is able to maintain smaller headways resulting in higher effective road capacity.

Similar calculations were carried out for the electric alternative. Many researchers delve into this

topic from the angle of marketability as Kang et al. (2016) touched on. Iacobucci et al. (2018)

proposed a unique concept that would allow SAEVs to save users money on their electricity bill.

The plan revolved around Virtual Power Plants (VPPs), a service popular in Japan that combines

many energy sources (with a focus on renewable sources) to supply power to consumers at

predetermined levels of demand. This predictable power load reduces stress on the grid and thereby

reduces the cost of electrical delivery. The study takes place in Tokyo and generates trips using

survey data. It is assumed that enough wind and solar generators can be built to cover all non-EV

electrical demand. Diesel generators would be available to keep up with excess electrical demand

from the SAEVs. The proposed VPP service would also include bundled SAEV service. The study

models scenarios with and without vehicle-to-grid (V2G) connectivity. V2G allows vehicles to

discharge power back to the grid to smooth power loads in times of high power demand. They

found major cost savings with their VPP, but those were dominated by the savings that come from

users not needing to own a car. Their system reduced electricity costs by 32% without V2G and

by 75% with V2G. SAEVs significantly increased carbon emissions in the study area. This is partly

because Japan’s current fleet of mostly hybrids is already relatively efficient, and secondly because

SAEVs introduce extra VMT due to empty travel.

Chen et al. (2016) also looked at the prices associated with EVs, vehicle automation, chargers,

insurance, maintenance, electricity, and replacement batteries. They found SAEVs would cost

about $0.60 to $0.67 per occupied mile to operate when including 10% profit margins. These

vehicles would be used an estimated 131 – 241 miles per day or 47,000 to 88,000 miles per year

assuming minimal downtime for weekends, maintenance, etc. This is compared to $0.58 per

occupied mile for non-EV SAVs with approximately 94,000 annual miles under the same

assumptions. Loeb and Kockelman (2019) followed up Chen et al.'s cost estimates with their more

detailed model and a few more cost considerations, such as cleaning, land acquisitions for charging

stations and demand-based electricity costs. They estimated a cost of about $0.59 per occupied

SAEV mile compared to $0.45 per occupied mile for a gasoline SAV fleet. This served to provide

a rigorous validation of Chen et al.’s estimates.

Kang et al. (2016) also looked carefully at a many of the same costs considered by Chen et al.

(2016) and Loeb and Kockelman (2019), like charger costs, vehicle acquisitions, electricity cost

and more. They used these figures to help understand SAVs’ level of competition with existing

car-sharing service Zipcar. They found a market-based profit-maximizing price of $2.50 per 10

minutes driving rate and a membership fee of $6 per month. Zipcar’s prices vary by location, but

as of 2019, a typical base membership is $6 - $7 per month and about $10 per hour ($1.67 per 10

minutes), up to 180 miles per day.

USER PERSPECTIVES

Research described above has supported the many benefits that SAVs can offer, with

recommendations on several different facets of SAV use. However, it is important to realize the

public perception of these advanced technologies and policies are adopted. This section summaries

research conducted on the users’ perspective on SAV use.

Bansal and Kockelman's (2018) study conducted in 2016 opened up the discussion on how users

perceive future SAV use. Their results show that SAV characteristics persuaded large-family

households and disabled users to be inclined towards SAV even when it was offered at $3.00 per

mile, which is comparable to present day taxis largely owing to lack of mobility options. However,

the average willingness to pay across all groups was low at $1.00 per mile. The characteristics of

those willing to use SAVs showed promise for future market penetration. Haboucha et al. (2017)

analyzed a survey that was distributed in Israel and North America on car-ownership and modeled

whether people are willing to choose shared vehicles over personal vehicles (both conventional

and autonomous). Their results, for the year 2015, observed a well-founded hesitation in switching

to AV technology. A majority of users willing to switch to SAVs was found to be younger, and

those who tend to spend more time in vehicles. Krueger et al.'s (2016) study focused on the choices

between SAVs requested for solo travel and SAVs with DRS. Their study showed that people

typically viewed these services as completely different options, with completely different values

of travel time, as expected and assumed in studies. Their results supported Haboucha et al.'s (2017)

finding that younger people are the most likely to adopt SAVs for frequent use.

More recent studies indicate that preferences have changed considerably over time. Quarles and

Kockelman (2018) took a closer look at how fares would impact respondents’ willingness to use

SAVs and DRS. Their results suggested a value of $0.44 per mile maximum willingness to pay

(WTP) to frequently use the service, which is much lower than what the service may cost (Bösch

et al., 2018a). Gurumurthy and Kockelman (2019) also surveyed people’s WTP for SAV and DRS

services in 2017. Their models predicted a WTP that was slightly higher on average (at $0.73 per

mile), although the share of respondents willing to use the service was low. An elasticity analysis

revealed that high usage levels were likely in the future with younger people inclined to use SAVs

and opt into DRS. More recently a survey by Stoiber et al. (2019) among Swiss households has

shown 61% of respondents prefer SAVs over personal-AVs. Perceptions may be continually

evolving, and these studies show that there may be a general trend towards SAV usage and DRS.

However, more surveys need to be conducted to capture this volatility in preferences.

DISCUSSION

This chapter covers the body of literature generated from 2010 to present in the realm of SAVs.

SAVs are almost here and researchers around the world have studied several aspects of SAV usage.

Table 1 shows a comprehensive list of articles relevant to this review laid out in chronological

order. The different aspects of SAVs covered so far in this chapter are listed across the table.

Several frameworks have been developed from scratch to analyze SAV use in different regions.

To date, the most widely applied model for SAV simulations (tracking travelers and vehicles)

appears to be MATSim. MATSim is open-source and able to run large-scale simulations. Recent

code contributions by researchers around the world have enabled simulation of DRS, SAEVs, and

advanced pricing policies. Another emerging agent-based modeling tool for SAVs is POLARIS

(Auld et al., 2016), as developed by the US Department of Energy’s Argonne National Laboratory.

Future Directions

In the near term, AVs and human-driven vehicles will share the same infrastructure. Some driver

apprehension may exist and these interactions need to be studied. Gaps in the rapidly evolving

SAV literature also exist for replacing or complementing traditional transit services, including

provision of first-mile and last-mile services. Large-vehicle SAVs with DRS is an important area

that can be explored further and use of SAVs for commercial applications is largely missing in the

literature. Other areas that merit further examination include congestion and parking impacts from

SAV services, in combination with strategic SAV-stop placement (for pickups and drop-offs,

especially in dense downtowns or other popular destinations), smart road tolling and microtolling

(using AVs’ smart GPS systems), and vehicle design, to facilitate relatively private travel during

DRS services, and route design, to enable deviations for relatively personalized aBus services. The

behavioral response to data privacy from connectivity and data generation in AVs may also impact

trip outcomes. These are some of the most immediate future research directions.

Table 1: Features and contributions of all relevant articles on SAVs

Relevant Articles (chronological) Region

Dyn

. R

ide-S

ha

re

Ele

ctri

c V

ehs.

Larg

e-S

cale

( >

1M

tri

ps)

Em

issi

on

s

Sim

ula

tio

n

Op

tim

izati

on

Toll

s/P

rici

ng

Cost

An

aly

sis

Use

r P

ersp

ecti

ve

Pa

rkin

g

Fir

st-M

ile-

Last

-Mil

e

Burns et al. (2013) 3 U.S. cities

Spieser et al. (2014) Singapore

Brownell and Kornhauser (2014) New Jersey

Fagnant and Kockelman (2014) 10 mi grid

Fagnant et al. (2015) Austin, TX

Burghout et al. (2015) Stockholm,

Sweden

Zhang et al. (2015) 10 mi grid

Bösch et al. (2016) Zurich,

Switzerland

Bischoff and Maciejewski (2016) Zurich,

Switzerland

Chen and Kockelman (2016) 100 mi grid

Haboucha et al. (2017)

Israel and

North

America

Krueger et al. (2016) Australia

Liang et al. (2016) Delft,

Netherlands

Chen et al. (2016) 100 mi grid

Kang et al. (2016) Ann Arbor,

MI

Fagnant and Kockelman (2018)

[published online in 2016] Austin, TX

Bansal and Kockelman (2018)

[published online in 2016] Texas

Maciejewski et al. (2017) -

Yang et al. (2017) Shanghai,

China

Hörl (2017) Sioux Falls,

SD

Heilig et al. (2017) Stuttgart,

Germany

Zhang and Guhathakurta (2017) Atlanta, GA

Martinez and Viegas (2017) Lisbon,

Portugal

Scheltes and de Almeida Correia

(2017)

Delft,

Netherlands

Levin et al. (2017) Austin, TX

Liu et al. (2017) Austin, TX

Sharon et al. (2017) -

Gurumurthy and Kockelman (2019)

[data from 2017] USA

Farhan et al. (2018) Seattle, WA

Hörl et al. (2018) Zurich,

Switzerland

Pinto et al. (2018) Evanston,

IL

Loeb et al. (2018) Austin, TX

Bösch et al. (2018a) Zurich,

Switzerland

Quarles and Kockelman (2018) USA

Hyland and Mahmassani (2018) Chicago, IL

Shen et al. (2018) Singapore

Iacobucci et al. (2018) Tokyo,

Japan

Farhan and Chen (2018) Austin, TX

Bösch et al. (2018b) Zurich,

Switzerland

Gurumurthy and Kockelman (2018) Orlando, FL

Ruch et al. (2018)

San

Francisco,

CA

Lu Miaojia et al. (2018) Ann Arbor,

MI

Zhao and Kockelman (2018) Austin, TX

Simoni et al. (2019) Austin, TX

Gurumurthy et al. (2019) Austin, TX

Lee and Kockelman (2019) Austin, TX

Stoiber et al. (2019) Switzerland

Loeb and Kockelman (2019) Austin, TX

Millard-Ball (2019)

San

Francisco,

CA

REFERENCES

Auld, J., Hope, M., Ley, H., Sokolov, V., Xu, B., Zhang, K., 2016. POLARIS: Agent-based

modeling framework development and implementation for integrated travel demand and

network and operations simulations. Transp. Res. Part C Emerg. Technol. 64, 101–116.

https://doi.org/10.1016/j.trc.2015.07.017

Bansal, P., Kockelman, K.M., 2018. Are we ready to embrace connected and self-driving

vehicles? A case study of Texans. Transportation 45, 641–675.

https://doi.org/10.1007/s11116-016-9745-z

Bischoff, J., Maciejewski, M., 2016. Simulation of City-wide Replacement of Private Cars with

Autonomous Taxis in Berlin. Procedia Comput. Sci. 83, 237–244.

https://doi.org/10.1016/j.procs.2016.04.121

Bösch, P.M., Becker, F., Becker, H., Axhausen, K.W., 2018a. Cost-based analysis of

autonomous mobility services. Transp. Policy 64, 76–91.

https://doi.org/10.1016/j.tranpol.2017.09.005

Bösch, P.M., Ciari, F., Axhausen, K.W., 2018b. Transport Policy Optimization with

Autonomous Vehicles. Transp. Res. Rec. 0361198118791391.

https://doi.org/10.1177/0361198118791391

Bösch, P.M., Ciari, F., Axhausen, K.W., 2016. Autonomous Vehicle Fleet Sizes Required to

Serve Different Levels of Demand. Transp. Res. Rec. 2542, 111–119.

https://doi.org/10.3141/2542-13

Brownell, C., Kornhauser, A., 2014. A Driverless Alternative: Fleet Size and Cost Requirements

for a Statewide Autonomous Taxi Network in New Jersey. Transp. Res. Rec. 2416, 73–

81. https://doi.org/10.3141/2416-09

Buehler, M., Iagnemma, K., Singh, S. (Eds.), 2009. The DARPA Urban Challenge: Autonomous

Vehicles in City Traffic, Springer Tracts in Advanced Robotics. Springer-Verlag, Berlin

Heidelberg.

Burghout, W., Rigole, P.-J., Andreasson, I., 2015. Impacts of Shared Autonomous Taxis in a

Metropolitan Area. Presented at the 94th Annual Meeting of the Transportation Research

Board.

Burns, L.D., Jordan, W.C., Scarborough, B.A., 2013. Transforming Personal Mobility 42.

Chen, T.D., Kockelman, K.M., 2016. Management of a Shared Autonomous Electric Vehicle

Fleet. Transp. Res. Rec. J. Transp. Res. Board 2572, 37–46. https://doi.org/10.3141/2572-

05

Chen, T.D., Kockelman, K.M., Hanna, J.P., 2016. Operations of a shared, autonomous, electric

vehicle fleet: Implications of vehicle & charging infrastructure decisions. Transp. Res.

Part Policy Pract. 94, 243–254. https://doi.org/10.1016/j.tra.2016.08.020

Clements, L.M., Kockelman, K.M., 2017. Economic Effects of Automated Vehicles. Transp.

Res. Rec. 2606, 106–114. https://doi.org/10.3141/2606-14

Fagnant, D.J., Kockelman, K., 2015. Preparing a nation for autonomous vehicles: opportunities,

barriers and policy recommendations. Transp. Res. Part Policy Pract. 77, 167–181.

https://doi.org/10.1016/j.tra.2015.04.003

Fagnant, D.J., Kockelman, K.M., 2018. Dynamic ride-sharing and fleet sizing for a system of

shared autonomous vehicles in Austin, Texas. Transportation 45, 143–158.

https://doi.org/10.1007/s11116-016-9729-z

Fagnant, D.J., Kockelman, K.M., 2014. The travel and environmental implications of shared

autonomous vehicles, using agent-based model scenarios. Transp. Res. Part C Emerg.

Technol. 40, 1–13. https://doi.org/10.1016/j.trc.2013.12.001

Fagnant, D.J., Kockelman, K.M., Bansal, P., 2015. Operations of Shared Autonomous Vehicle

Fleet for Austin, Texas, Market. Transp. Res. Rec. J. Transp. Res. Board 2536, 98–106.

https://doi.org/10.3141/2536-12

Farhan, J., Chen, T.D., 2018. Impact of ridesharing on operational efficiency of shared

autonomous electric vehicle fleet. Transp. Res. Part C Emerg. Technol. 93, 310–321.

https://doi.org/10.1016/j.trc.2018.04.022

Farhan, J., Chen, T.D., Zhang, Z., 2018. Leveraging Shared Autonomous Electric Vehicles for

First/Last Mile Mobility. Presented at the 97th Annual Meeting of the Transportation

Research Board, Washington, D.C.

Gurumurthy, K.M., Kockelman, K.M., 2019. Modeling Americans’ Autonomous Vehicle

Preferences: A Focus on Dynamic Ride-Sharing, Privacy & Long-Distance Mode

Choices. Presented at the 98th Annual Meeting of the Transportation Research Board,

Washington, D.C.

Gurumurthy, K.M., Kockelman, K.M., 2018. Analyzing the dynamic ride-sharing potential for

shared autonomous vehicle fleets using cellphone data from Orlando, Florida. Comput.

Environ. Urban Syst. 71, 177–185. https://doi.org/10.1016/j.compenvurbsys.2018.05.008

Gurumurthy, K.M., Kockelman, K.M., Simoni, M.D., 2019. Benefits and Costs of Ride-Sharing

in Shared Automated Vehicles across Austin, Texas: Opportunities for Congestion

Pricing. Transp. Res. Rec. 2673, 548–556. https://doi.org/10.1177/0361198119850785

Haboucha, C.J., Ishaq, R., Shiftan, Y., 2017. User preferences regarding autonomous vehicles.

Transp. Res. Part C Emerg. Technol. 78, 37–49. https://doi.org/10.1016/j.trc.2017.01.010

Heilig, M., Hilgert, T., Mallig, N., Kagerbauer, M., Vortisch, P., 2017. Potentials of Autonomous

Vehicles in a Changing Private Transportation System – a Case Study in the Stuttgart

Region. Transp. Res. Procedia, Emerging technologies and models for transport and

mobility 26, 13–21. https://doi.org/10.1016/j.trpro.2017.07.004

Hörl, S., 2017. Agent-based simulation of autonomous taxi services with dynamic demand

responses. Procedia Comput. Sci. 109, 899–904.

https://doi.org/10.1016/j.procs.2017.05.418

Hörl, S., Ruch, C., Becker, F., Frazzoli, E., Axhausen, K.W., 2018. Fleet control algorithms for

automated mobility: A simulation assessment for Zurich. Presented at the 97th Annual

Meeting of the Transportation Research Board.

Horni, A., Nagel, K., Axhausen, K.W. (Eds.), 2016. The Multi-Agent Transport Simulation

MATSim. Ubiquity Press. https://doi.org/10.5334/baw

Hyland, M., Mahmassani, H.S., 2018. Dynamic autonomous vehicle fleet operations:

Optimization-based strategies to assign AVs to immediate traveler demand requests.

Transp. Res. Part C Emerg. Technol. 92, 278–297.

https://doi.org/10.1016/j.trc.2018.05.003

Iacobucci, R., McLellan, B., Tezuka, T., 2018. The Synergies of Shared Autonomous Electric

Vehicles with Renewable Energy in a Virtual Power Plant and Microgrid. Energies 11,

2016. https://doi.org/10.3390/en11082016

Kang, N., Feinberg, F.M., Papalambros, P.Y., 2016. Autonomous Electric Vehicle Sharing

System Design. J. Mech. Des. 139, 011402-011402–10.

https://doi.org/10.1115/1.4034471

Kockelman, K., Boyles, S. (Eds.), 2018. Anticipating Emissions Impacts of Smoother Driving by

Connected and Autonomous Vehicles Using MOVES Model, in: Smart Transport for

Cities & Nations: The Rise of Self-Driving & Connected Vehicles.

Kockelman, K., Boyles, S., Avery, P., Claudel, C., Loftus-Otway, L., Fagnant, D., Bansal, P.,

Levin, M.W., Zhao, Y., Liu, J., Clements, L., Wagner, W., Stewart, D., Sharon, G.,

Albert, M., Stone, P., Hanna, J., Patel, R., Fritz, H., Choudhary, T., Li, T., Nichols, A.,

Sharma, K., Simoni, M., 2016. Bringing Smart Transport to Texans: Ensuring the

Benefits of a Connected and Autonomous Transport System in Texas (No. FHWA/TX-

16/0-6838-2). TxDOT.

Kockelman, K., Boyles, S., Claudel, C., Stone, P., Loftus-Otway, L., Sturgeon, P., Sharon, G.,

Gurumurthy, K.M., Huang, Y., Simoni, M., Lei, T., Patel, R., He, D., Mohamed, A., Liu,

J., Yarmohammadi, S., Thorn, E., Wagner, W., Stewart, D., Albert, M., Hanna, J., 2018.

Phase 2 - Bringing Smart Transport to Texans: Ensuring the Benefits of a Connected and

Autonomous Transport System in Texas (No. FHWA/TX-18/0-6838-3). TxDOT.

Krueger, R., Rashidi, T.H., Rose, J.M., 2016. Preferences for shared autonomous vehicles.

Transp. Res. Part C Emerg. Technol. 69, 343–355.

https://doi.org/10.1016/j.trc.2016.06.015

Lee, J., Kockelman, K.M., 2019. Energy and Emissions Implications of Self-Driving Vehicles.

Presented at the 98th Annual Meeting of the Transportation Research Board,

Washington, D.C.

Levin, M.W., Kockelman, K.M., Boyles, S.D., Li, T., 2017. A general framework for modeling

shared autonomous vehicles with dynamic network-loading and dynamic ride-sharing

application. Comput. Environ. Urban Syst. 64, 373–383.

https://doi.org/10.1016/j.compenvurbsys.2017.04.006

Liang, X., Correia, G.H. de A., van Arem, B., 2016. Optimizing the service area and trip

selection of an electric automated taxi system used for the last mile of train trips. Transp.

Res. Part E Logist. Transp. Rev. 93, 115–129. https://doi.org/10.1016/j.tre.2016.05.006

Liu, J., Kockelman, K.M., Bösch, P.M., Ciari, F., 2017. Tracking a system of shared autonomous

vehicles across the Austin, Texas network using agent-based simulation. Transportation

44, 1261–1278. https://doi.org/10.1007/s11116-017-9811-1

Loeb, B., Kockelman, K.M., 2019. Fleet performance and cost evaluation of a shared

autonomous electric vehicle (SAEV) fleet: A case study for Austin, Texas. Transp. Res.

Part Policy Pract. 121, 374–385. https://doi.org/10.1016/j.tra.2019.01.025

Loeb, B., Kockelman, K.M., Liu, J., 2018. Shared autonomous electric vehicle (SAEV)

operations across the Austin, Texas network with charging infrastructure decisions.

Transp. Res. Part C Emerg. Technol. 89, 222–233.

https://doi.org/10.1016/j.trc.2018.01.019

Lu Miaojia, Taiebat Morteza, Xu Ming, Hsu Shu-Chien, 2018. Multiagent Spatial Simulation of

Autonomous Taxis for Urban Commute: Travel Economics and Environmental Impacts.

J. Urban Plan. Dev. 144, 04018033. https://doi.org/10.1061/(ASCE)UP.1943-

5444.0000469

Maciejewski, M., Bischoff, J., Hörl, S., Nagel, K., 2017. Towards a Testbed for Dynamic

Vehicle Routing Algorithms, in: Bajo, J., Vale, Z., Hallenborg, K., Rocha, A.P., Mathieu,

P., Pawlewski, P., Del Val, E., Novais, P., Lopes, F., Duque Méndez, N.D., Julián, V.,

Holmgren, J. (Eds.), Highlights of Practical Applications of Cyber-Physical Multi-Agent

Systems, Communications in Computer and Information Science. Springer International

Publishing, pp. 69–79.

Martinez, L.M., Viegas, J.M., 2017. Assessing the impacts of deploying a shared self-driving

urban mobility system: An agent-based model applied to the city of Lisbon, Portugal. Int.

J. Transp. Sci. Technol., Connected and Automated Vehicles: Effects on Traffic,

Mobility and Urban Design 6, 13–27. https://doi.org/10.1016/j.ijtst.2017.05.005

Millard-Ball, A., 2019. The autonomous vehicle parking problem. Transp. Policy 75, 99–108.

https://doi.org/10.1016/j.tranpol.2019.01.003

Pinto, H.K.R. de F., Hyland, M.F., Verbas, İ.Ö., Mahmassani, H.S., 2018. Integrated Mode

Choice and Dynamic Traveler Assignment-Simulation Framework to Assess the Impact

of a Suburban First-Mile Shared Autonomous Vehicle Fleet Service on Transit Demand.

Presented at the 97th Annual Meeting of the Transportation Research Board,

Washington, D.C.

Quarles, N.T., Kockelman, K.M., 2018. Americans’ Plans for Acquiring and Using Electric,

Shared and Self-Driving Vehicles.

Ruch, C., Hörl, S., Frazzoli, E., 2018. AMoDeus, a Simulation-Based Testbed for Autonomous

Mobility-on-Demand Systems, in: 2018 21st International Conference on Intelligent

Transportation Systems (ITSC). Presented at the 2018 21st International Conference on

Intelligent Transportation Systems (ITSC), pp. 3639–3644.

https://doi.org/10.1109/ITSC.2018.8569961

Scheltes, A., de Almeida Correia, G.H., 2017. Exploring the use of automated vehicles as last

mile connection of train trips through an agent-based simulation model: An application to

Delft, Netherlands. Int. J. Transp. Sci. Technol., Connected and Automated Vehicles:

Effects on Traffic, Mobility and Urban Design 6, 28–41.

https://doi.org/10.1016/j.ijtst.2017.05.004

Sharon, G., Levin, M.W., Hanna, J.P., Rambha, T., Boyles, S.D., Stone, P., 2017. Network-wide

adaptive tolling for connected and automated vehicles. Transp. Res. Part C Emerg.

Technol. 84, 142–157. https://doi.org/10.1016/j.trc.2017.08.019

Shen, Y., Zhang, H., Zhao, J., 2018. Integrating shared autonomous vehicle in public

transportation system: A supply-side simulation of the first-mile service in Singapore.

Transp. Res. Part Policy Pract. 113, 125–136. https://doi.org/10.1016/j.tra.2018.04.004

Simoni, M.D., Kockelman, K.M., Gurumurthy, K.M., Bischoff, J., 2019. Congestion pricing in a

world of self-driving vehicles: An analysis of different strategies in alternative future

scenarios. Transp. Res. Part C Emerg. Technol. 98, 167–185.

https://doi.org/10.1016/j.trc.2018.11.002

Spieser, K., Treleaven, K., Zhang, R., Frazzoli, E., Morton, D., Pavone, M., 2014. Toward a

Systematic Approach to the Design and Evaluation of Automated Mobility-on-Demand

Systems: A Case Study in Singapore, in: Meyer, G., Beiker, S. (Eds.), Road Vehicle

Automation, Lecture Notes in Mobility. Springer International Publishing, Cham, pp.

229–245. https://doi.org/10.1007/978-3-319-05990-7_20

Stoiber, T., Schubert, I., Hoerler, R., Burger, P., 2019. Will consumers prefer shared and pooled-

use autonomous vehicles? A stated choice experiment with Swiss households. Transp.

Res. Part Transp. Environ. https://doi.org/10.1016/j.trd.2018.12.019

Yang, J., Dong, J., Wang, W., 2017. A Simulation Model for Performance Analysis of Electric

Autonomous Taxi Systems. Presented at the 96th Annual Meeting of the Transportation

Research Board.

Zhang, W., Guhathakurta, S., 2017. Parking Spaces in the Age of Shared Autonomous Vehicles.

Transp. Res. Rec. J. Transp. Res. Board 2651, 80–91. https://doi.org/10.3141/2651-09

Zhang, W., Guhathakurta, S., Fang, J., Zhang, G., 2015. Exploring the impact of shared

autonomous vehicles on urban parking demand: An agent-based simulation approach.

Sustain. Cities Soc. 19, 34–45. https://doi.org/10.1016/j.scs.2015.07.006

Zhao, Y., Kockelman, K.M., 2018. Anticipating the Regional Impacts of Connected and

Automated Vehicle Travel in Austin, Texas. J. Urban Plan. Dev. 144, 04018032.

https://doi.org/10.1061/(ASCE)UP.1943-5444.0000463

GLOSSARY

aBus Autonomous Bus

AV Autonomous Vehicle

AVO Average Vehicle Occupancy

DRS Dynamic Ride-Sharing

DTA Dynamic Traffic Assignment

EV Electric Vehicle

ICE Internal Combustion Engine

MATSim Multi-Agent Transport Simulation

O-D Origin-Destination

OEM Original Equipment Manufacturer

SAEV Shared Autonomous Electric Vehicle

SAV Shared Autonomous Vehicle

TNC Transportation Network Company

V2G Vehicle-to-Grid

VMT Vehicle-Miles Traveled

WTP Willingness to Pay